Bayesian Anytime m-top Exploration

Pieter Libin; Timothy Verstraeten; Diederik M Roijers; Wenjia Wang; Kristof Theys; Ann Nowe

doi:10.1109/ICTAI.2019.00201

Bayesian Anytime m-top Exploration

Pieter Libin, Timothy Verstraeten, Diederik M Roijers, Wenjia Wang, Kristof Theys, Ann Nowe

Research output: Chapter in Book/Report/Conference proceeding › Conference paper

2 Citations (Scopus)

323 Downloads (Pure)

Abstract

We introduce Boundary Focused Thompson sampling (BFTS), a new Bayesian algorithm to solve the anytime m-top exploration problem, where the objective is to identify the m best arms in a multi-armed bandit. First, we consider a set of existing benchmark problems that consider sub-Gaussian reward distributions (i.e., Gaussian with fixed variance and categorical reward). Next, we introduce a new environment inspired by a real world decision problem concerning insect control for organic agriculture. This new environment encodes a Poisson rewards distribution. For all these benchmarks, we experimentally show that BFTS consistently outperforms AT-LUCB, the current state of the art algorithm.

Original language	English
Title of host publication	2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI)
Pages	1422-1428
Number of pages	7
ISBN (Electronic)	2375-0197
DOIs	https://doi.org/10.1109/ICTAI.2019.00201
Publication status	Published - Nov 2019
Event	ICTAI - Portland, United States Duration: 4 Nov 2019 → 6 Nov 2019

Conference

Conference	ICTAI
Country/Territory	United States
City	Portland
Period	4/11/19 → 6/11/19

Access to Document

10.1109/ICTAI.2019.00201

bfts-pieter.libinAccepted author manuscript, 680 KB

Handle.net

Persistent link

Cite this

@inproceedings{f2836ffe52204c32b17f50649f4f0964,

title = "Bayesian Anytime m-top Exploration",

abstract = "We introduce Boundary Focused Thompson sampling (BFTS), a new Bayesian algorithm to solve the anytime m-top exploration problem, where the objective is to identify the m best arms in a multi-armed bandit. First, we consider a set of existing benchmark problems that consider sub-Gaussian reward distributions (i.e., Gaussian with fixed variance and categorical reward). Next, we introduce a new environment inspired by a real world decision problem concerning insect control for organic agriculture. This new environment encodes a Poisson rewards distribution. For all these benchmarks, we experimentally show that BFTS consistently outperforms AT-LUCB, the current state of the art algorithm.",

author = "Pieter Libin and Timothy Verstraeten and Roijers, {Diederik M} and Wenjia Wang and Kristof Theys and Ann Nowe",

year = "2019",

month = nov,

doi = "10.1109/ICTAI.2019.00201",

language = "English",

isbn = "978-1-7281-3798-8",

pages = "1422--1428",

booktitle = "2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI)",

note = "ICTAI ; Conference date: 04-11-2019 Through 06-11-2019",

}

TY - GEN

T1 - Bayesian Anytime m-top Exploration

AU - Libin, Pieter

AU - Verstraeten, Timothy

AU - Roijers, Diederik M

AU - Wang, Wenjia

AU - Theys, Kristof

AU - Nowe, Ann

PY - 2019/11

Y1 - 2019/11

N2 - We introduce Boundary Focused Thompson sampling (BFTS), a new Bayesian algorithm to solve the anytime m-top exploration problem, where the objective is to identify the m best arms in a multi-armed bandit. First, we consider a set of existing benchmark problems that consider sub-Gaussian reward distributions (i.e., Gaussian with fixed variance and categorical reward). Next, we introduce a new environment inspired by a real world decision problem concerning insect control for organic agriculture. This new environment encodes a Poisson rewards distribution. For all these benchmarks, we experimentally show that BFTS consistently outperforms AT-LUCB, the current state of the art algorithm.

AB - We introduce Boundary Focused Thompson sampling (BFTS), a new Bayesian algorithm to solve the anytime m-top exploration problem, where the objective is to identify the m best arms in a multi-armed bandit. First, we consider a set of existing benchmark problems that consider sub-Gaussian reward distributions (i.e., Gaussian with fixed variance and categorical reward). Next, we introduce a new environment inspired by a real world decision problem concerning insect control for organic agriculture. This new environment encodes a Poisson rewards distribution. For all these benchmarks, we experimentally show that BFTS consistently outperforms AT-LUCB, the current state of the art algorithm.

U2 - 10.1109/ICTAI.2019.00201

DO - 10.1109/ICTAI.2019.00201

M3 - Conference paper

SN - 978-1-7281-3798-8

SP - 1422

EP - 1428

BT - 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI)

T2 - ICTAI

Y2 - 4 November 2019 through 6 November 2019

ER -

Bayesian Anytime m-top Exploration

Abstract

Conference

Access to Document

Handle.net

Fingerprint

Cite this